Inter-subject variability of multi-organ shapes and their interrelations are modeled and applied to automated segmentation from CT images. In order to deal with large inter-subject variability such as that of the musculoskeletal system and abdomen, hierarchical modeling and conditional modeling methods based on interrelations among organ are introduced. The proposed methods are shown to be effective for segmentation of the musculoskeletal structures of the hip joint including the pelvis, femur, articular cartilage, and muscles, and the upper abdominal organs including the liver, kidneys, spleen, pancreas, gallbladder, aorta, inferior vena cava, and GI-tract. Assuming that automated organ segmentation is possible, the obtained organ shape information is further utilized for statistical modeling of diagnostic and therapeutic decision support. We develop a method for statistical modeling of the disease-specific shape components and show its usefulness in liver fibrosis diagnostic assistance. In addition, we develop methods for statistical modeling of the implants and anatomical structures, which are shown to be effective in automated surgical planning of total hip arthroplasty.
View full abstract